Typical g-C3N4 catalyst (denoted CN) was prepared by sintering melamine powder at 550°C in a muffle furnace according to a classic protocol from literature (for details of the preparation method, see Supplementary Information)17. We first used isotopic16O/18O labelled H2O for in-situ tracing of possible OER intermediate at the H2O/CN interface during continuous reaction by DRIFTS. H2O molecules were carried into the reaction chamber by N2 flow until equilibrium. Setting the equilibrium condition as the blank background, positive or negative IR response signal directly reflects the gain or loss of intermediate species at H2O/catalyst interface. As shown in Fig. 1a, when the CN/H2O sample was used and irradiated in situ with a 420 nm LED lamp, a broad negative absorption band from 3700 cm− 1 to 3000 cm− 1 and a very weak negative peak at 1645 cm− 1 emerged from the background and increased in intensity with increasing irradiation time. The broad negative band was assigned to the stretching vibration of O-H bond, whereas the weak negative peak at 1645 cm− 1 was from the bending vibration of H-O-H of H2O molecules, representing the loss of surface -OH species and H2O molecules during continuous OER18,19. The signal of O-H stretching vibration was much larger than the bending vibration signal of H2O molecules, suggesting that the OER occurred at H2O/CN interface was predominantly in the form of dissociated O-H. Identical features were observed when CN was replaced with other CN sample (i.e., F0.1-CN), or when H2O was replaced with 18O-labelled H218O (Fig. 1a-1d) as the signature of OER at the H2O/CN interface. More importantly, an increasing positive peak at 1725 cm− 1 that ascribed to the C = O stretching vibration was observed with increasing irradiation time (Fig. 1a), indicating that the collective formation of C = O species on CN surface. When we replaced H2O with 18O-labelled H218O under otherwise identical conditions (Fig. 1b), the positive peak at 1725 cm− 1 and the newly generated peak at 1524 cm− 1 emerged in terms of the theoretical 16O/18O replacement effect, which confirms that the O source of C = O was from H2O and further provides direct evidence for C = O formation during photocatalytic OER at H2O/CN interface. Such a C = O formation can only occur with carbon sites on CN being oxidized. To prevent the formation of C = O on CN, we devised a surface fluorination strategy to occupy carbon-sites on CN with F− ions through a hydrothermal treatment. Prepared fluorinated CN samples (denoted F-CN) were labeled as F0.01-CN ~ F1-CN with different F− concentration (0.01 mM ~ 1 mM) of the precursor for fluorination (for details of the preparation method, see Supplementary Information). The surface fluorination did not severely change the morphology (Supplementary Fig. 1) and crystalline structure of CN (Supplementary Fig. 2), but formed a strong C-F interaction (Supplementary Fig. 3). When we replaced CN with F0.1-CN (Fig. 1c), the positive C = O signal was no longer observed at the H2O/F0.1-CN interface. Further replacing the H2O with 18O-labelled H218O under otherwise identical conditions showed neither C = O nor C = 18O diagnostic signals (Fig. 1d), which solidly confirms that the fluorination of CN prevents the carbon sites being oxidized into C = O intermediates.
The collective formation of C = O state by oxidizing carbon sites on CN was also directly observed by NAP-XPS. The NAP-XPS spectra were in-situ collected in a vacuum chamber with 0.2 mbar H2O vapor pressure. A 300 W Xenon lamp as the white light source was placed outside the chamber to illuminate the sample via the quartz window. On the O1s spectra of the pristine CN sample, two major peaks at 530.1 eV and 531.3 eV were observed, corresponding to oxygen states of C-O and O-H species (Fig. 1e), respectively20. Under the white light illumination, a newly emerged contribution at 532.7 eV from C = O configuration was observed and increased in intensity with increasing irradiation time (Fig. 1e). Moreover, on the C1s spectra, peaks of C-C and N = C-N states on the pristine CN were gradually shifted towards higher binding energy from 284.4 eV and 287.7 eV to 284.9 eV and 288.1 eV (Supplementary Fig. 4a), respectively, under continuous white light illumination, corresponding to the formation of an oxidized carbon state on CN21. The NAP-XPS result is consistent with in-situ DRIFTS observations (Fig. 1a,b), demonstrating that C = O intermediate state was indeed formed at the H2O/CN interface during OER. After fluorination, although the strong C-F interaction can be recognized from the C1s peak shifting of F0.1-CN sample in comparison with CN (Supplementary Fig. 3a), little changes were found on both O1s (Fig. 1f) and C1s Supplementary Fig. 4b) spectra of F0.1-CN during continuous white light illumination with 0.2 mbar H2O vapor, which further demonstrates that C = O formation was vastly minimized on F-CN. In contrast, no changes were found on the N1s spectra on both CN and F0.1-CN samples (Supplementary Fig. 4c,d).
We argue that the strong C = O bonding from carbon site oxidation is an inherent bottleneck for OER on single-phased CN catalysts. If that is the case, preventing the intermediate C = O formation would endow CN catalysts deserved overall water splitting performances. Photocatalytic overall water splitting experiments on CN and different F-CN samples were performed in pure water without any organic sacrificial reagents under both the white light (Fig. 2a) and AM1.5G simulated solar irradiation (Fig. 2b). Under continuous white light irradiation (Xe lamb, 1000 mW cm− 2), the pristine CN catalyst only exhibited a mild H2 evolution of 11.60 µmol∙g− 1∙h− 1 without O2 evolution. After hydrothermal treatment, CN exfoliated thin layer sample (denoted CN-E) showed a slightly higher H2 evolution rate of 20.22 µmol∙g− 1∙h− 1 due to the enlarged specific surface area of CN-E (62.12 m2∙g− 1) in comparison with CN (8.66 m2∙g− 1), but still no O2 evolution observed. The poor performance on CN and CN-E catalysts is consistent with literature reports22,23, demonstrating that single-phased g-C3N4 catalyst dose not possess the overall water splitting ability. However, after the fluorination treatment, all F-CN catalysts exhibited both H2 and O2 evolution capabilities under identical experimental conditions, which varies with the fluorination degree. Particularly, the champion F0.1-CN catalyst exhibited the H2 evolution rate of 174.77 µmol∙g− 1∙h− 1, which is 15.06 and 8.64 times higher than those of the pristine CN and CN-E catalysts, respectively, and continuous O2 evolution of 44.15 µmol∙g− 1∙h− 1 (Fig. 2a). Although the specific surface area of F0.1-CN (42.69 m2∙g− 1) is larger than that of the pristine CN (8.66 m2∙g− 1) after hydrothermal exfoliation treatment, it is still smaller than that of CN-E (62.12 m2∙g− 1) (Supplementary Fig. 5), yet F0.1-CN exhibited an order-of-magnitude-improved water splitting efficiency (Fig. 2a,b). Further increase the F− ions would slightly decrease the performance of as-prepared F-CN catalysts, which is attributed to the enhanced hydrophobic feature by fluorination (Supplementary Fig. 6). Moreover, under AM1.5 simulated solar irradiation, the F0.1-CN catalyst still exhibited excellent overall water splitting capacity with H2 evolution rate of 83.89 µmol∙g− 1∙h− 1, increasing by 9.63 times in comparison with the pristine CN catalyst (8.71 µmol∙g− 1∙h− 1), and continuous O2 evolution rate of 21.15 µmol∙g− 1∙h− 1. Control experiments have been done to confirm no H2/O2 productions were detected in the dark, no catalysts or without H2O for the F0.1-CN catalyst (Supplementary Fig. 7). Isotopic labeled experiments also confirmed that H2 and O2 were produced sorely from the photocatalytic water splitting rather than other effects, whereas D2 and 18O2 were detected as products of D2O and 18O-labelled H218O (Supplementary Fig. 8). Notably, H2/O2 production ratio on F-CN catalysts was less than the stoichiometric ratio of 2:1 (Fig. 2a,b). The short of O2 production on F-CN was due to the further reduction of O2 into H2O2, since CN is very active for O2 reduction24,25, which was further demonstrated by the in-situ observation of H2O2 production during the reaction (Supplementary Fig. 9) by the well-reported Ghormley triiodide method26. Furthermore, within 20 hours, F0.1-CN can still maintain more than 80% of efficiency on H2 and O2 production and continue to work, in contrast, CN and CN-E were quickly deactivated with less than 50% of initial efficiency on H2 production within 8 hours (Supplementary Fig. 10), indicating that the H2 evolution on F0.1-CN came from the continuous overall water splitting, whereas the mild H2 evolution on CN and CN-E was possibly from the unsustainable self-oxidation.
Comparing above photocatalytic performance results with our in-situ DRIFTS (Fig. 1a-1d) and in-situ NAP-XPS (Fig. 1e,f) observations, we reason that the intermediate C = O formation is the bottleneck of overall water splitting on single-phased CN catalysts. To further verify that the emerging overall water splitting ability on F-CN is due to the preventing of C = O formation rather than other effects, we first compared the visible-light absorption of CN and F-CN catalysts. Figure 2c shows the wavelength dependence of apparent quantum yield (AQY) on the pristine CN and the champion F0.1-CN catalysts along with the ultraviolet-visible diffuse reflection spectra (UV-vis DRS). As peak values, AQYs at 365 nm on both samples were determined to be 0.5718% (F0.1-CN) and 0.1281% (CN). When the incident wavelength increased from 365 nm to 500 nm, AQYs of both samples were sharply decreased (Supplementary Table 1 and Table 2), which coincides with literature reports on g-C3N4-based catalysts27,28. In visible-light region, AQYs at 420 nm on both samples were determined to be 0.0164% (F0.1-CN) and 0.0005% (CN). The much higher AQYs on F0.1-CN catalyst than that on the pristine CN catalyst further evince the effect of fluorination treatment. However, from UV-vis DRS spectra, no discernible differences on the absorption edge between CN and F0.1-CN were observed, indicating that the improved overall water splitting performance of F-CN was not from the enhanced visible-light response.
We further tracked the transient fluorescence emission profile at 465 nm on both F0.1-CN and CN catalysts with the incident 375 nm irradiation and found that the emission lifetime of the F0.1-CN catalyst is not significantly extended in comparison with the pristine CN catalyst (Fig. 2d). The fitted emission decay profiles suggest a slightly shortened exciton lifetime on F0.1-CN with lifetime parameter reduced from τA = 7.72 ns to τA = 6.36 ns in comparison with the pristine CN, which denies the extended exciton lifetime as the major effect of fluorination for enhanced overall water splitting performances. Moreover, from the XPS valence band (VB) spectra near Fermi level (Supplementary Fig. 11), both CN and F0.1-CN exhibited almost identical VB position at 1.88 eV, which denies the VB position as the major contributor for the order-of-magnitude performance improvement on F-CN.
Through above characterizations, we ruled out that the morphology, crystalline structure, visible-light response, exciton lifetimes, and VB positions are the main factors affecting the performance of F-CN for overall water splitting. However, by using in-situ DRIFTS (Fig. 1a-1d) and NAP-XPS (Fig. 1e,f, Supplementary Fig. 4), we successfully identified the formation of C = O intermediate and their minimization on F-CN surface, which is completely consistent with the tendency of photocatalytic water splitting performances. Therefore, we conclude that the formation of C = O intermediate is an important bottleneck for overall water splitting on single-phased CN catalyst. By occupying the carbon site on CN surface by fluorination, this bottleneck can be bypassed to achieve efficient visible-light-driven H2 and O2 productions from overall water splitting on single-phased F-CN catalysts.
DFT calculations were further employed to investigate the effect of surface flurination on the water decomposition reaction (i.e., OER) on CN/F-CN surface (for details of the computational methods, see Supplementary Information). F-CN layer was formed by using one F atom to bond with the C atom in CN (Supplementary Fig. 12). Water adsorption and activation were simulated on both C sites (Fig. 3a) and N sites (Supplementary Fig. 13a) in pristine CN and N atoms adjacent to the C-F bond in F-CN as reactive sites (Fig. 3b). Calculated free energy profiles show that OER on surface C site has a lower energy barrier (2.25 eV) than the surface N site (2.86 eV) in CN (Fig. 3c), indicating that C site is the predominant OER reactive sites in the pristine CN. Moreover, after F atom ocupying the C site, the surface N site in F-CN owns a much lower energy barrier (1.58 eV) than the surface C (2.25 eV) or N (2.86 eV) site in pristine CN (Fig. 3c), demonstrating that the F modification indeed can improve the OER activity on F-CN. According to the corresponding evolution pathway of geometry structures (Fig. 3a,b), the improved OER activity is considered due to the C-F bond formation optimizing the OER pathway on adjacent N atoms. Especially, the calculated charge density difference mappings show that the F modification in F-CN optimizes the bonding interaction between CN surface and *OH intermediate (Fig. 3e), which effectively avoids the excessively strong C-O interaction (Fig. 3d) or weak N-O interaction (Supplementary Fig. 13b) in pristine CN. As a result, the F modification greatly decreases the formation energy of rate-determining *OH. It should be noted that the F modification also significantly promote the formation of *OOH, which is also a high-barrier reaction step in OER (Fig. 3c). This implies that the excessively stable *O intermediate in the form of C = O bond on pristine CN is difficult to be further converted into *OOH. As a result, CN with an observable IR signal of C = O during reaction owns a lower activity than F-CN, which completely coincides with our experimental observations. Furthermore, the PDOS reveals that the different bonding behavior between *OH and catalyst surface is attributed to the F modification that enables the N2p states to move upward the Fermi level (Fig. 3f), which effectively enhances the stability of formed *OH on the N site in F-CN. Thus, the N site in F-CN is the main OER center.